mgm (version 1.1-7)

tv.mgmsampler:

Description

Samples from a time-varying pairwise Mixed Graphical Model

Usage

tv.mgmsampler(type, lev, graphs, threshs, parmatrices = NA, nIter = 250, varadj = .2)

Arguments

type

p (variables) x 1 vector specifying the type of distribution for each variable ("g" = Gaussian, "p" = Poisson, "e" = Exponential, "c" = Categorical)

lev

p x 1 vector specifying the number of levels for each variables (for continuous variables = 1)

graphs

A p x p x n (time steps) symmetric (weighted) adjacency matrix, where n are the number of time steps, i.e. a graph is specified for each time step.

threshs

A list in which each entry corresponds to a time step. For each time step there is a list in which each entry corresponds to one variable (as thresh in mgmsampler.)

parmatrices

Optional; Equivalent to parmatrix in mgmsampler, only with a parameter matrix provided for each time step n. If provided, graphs will be ignored.

nIter

Number of iterations in the Gibbs sampler

varadj

Additive constant to conditional Gaussian variances before normalization. This avoids partial correlations close to 1.

Value

n x p data matrix.

References

Haslbeck, J., & Waldorp, L. J. (2015). Structure estimation for mixed graphical models in high-dimensional data. arXiv preprint arXiv:1510.06871.

Yang, E., Baker, Y., Ravikumar, P., Allen, G., & Liu, Z. (2014). Mixed graphical models via exponential families. In Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics (pp. 1042-1050).

Wainwright, M. J., & Jordan, M. I. (2008). Graphical models, exponential families, and variational inference. Foundations and Trends in Machine Learning, 1(1-2), 1-305.

Examples

## Not run: ------------------------------------
# # We sample from a graph with 400 time steps.
# # In the first 200 steps the underlying graph is graph A, 
# # and in the last 200 steps graph B
# 
#   p <- 4 # 4 variables
#   n <- 400 # number of time steps
#   graphA <- graphB <- matrix(0, p, p)
#   graphA[2,1] <- graphA[1,2] <- 1 # graphA has edge (2,1)
#   graphB[3,4] <- graphB[4,3] <- 1 # graphB has edge (3,4)
#   graphs <- array(dim=c(p, p, n)) 
#   graphs[,,1:(n/2)] <- graphA
#   graphs[,,(n/2+1):n] <- graphB
#   
#   # specify type and thresholds
#   type <- c('g', 'g', 'c', 'c') # two Gaussian, two Binary
#   lev <- c(1, 1, 2, 2)
#   # zero means (Gaussians) and thresholds (Binary)
#   thresh <- list(0, 0 , c(0, 0), c(0, 0))
#   # same means/thresholds at all time steps
#   threshs <- list(); for(nIter in 1:n) threshs[[nIter]] <- thresh
# 
#   # sample
#   set.seed(1)
#   data <- tv.mgmsampler(type, lev, graphs, threshs)
#   
#   # look at sampled data
#   dim(data)
#   head(data)
## ---------------------------------------------